I was at a MSFT event for the last 2 days on future VS.NET / .NET FX stuff (I’ll avoid the pre-MSFT-Scoble-esque Longhorn-rocks-but-I-can’t-talk-about-it-because-it’s-NDA posts). I ran into a bunch of Software Legends who all knew me.

• Don Box, who made a big deal of remembering my name.
• Chris Sells, who knew me from my Messenger.NET stuff I built on his imcli classes.
• Jeffrey Richter, who doesn’t work for MSFT like Chris and Don, but has been consulting w/ the VS.NET team since early in the .NET FX dev cycle. I took an early class on .NET (Dec 2000) from Jeff.

This is part of why I love working for MSFT. You add Paul DiLascia (author of Windows++) to that mix and you would have the core group of authors that helped me learn Win32, MFC, COM and ATL. I doubt my previous jobs would have afforded me this opportunity.

The prescriptive guidance is coming fast and furious now. The User Interface Process Application Block is a framework for abstracting the control flow and state management out of the user interface layer. The Configuration Management Application Block is a solution to manage configuration data across your applications. And the Updater Application Block provides a mechanism to build the ability to update directly into your application.

Also, from ScottW, I discovered that the Data Access Application Block v2 shipped last Friday. Hasn’t shown up on MSDN yet.

Got a chance to talk to Dare today about the XML vs. Objects stuff we’ve been blogging on. We started by talking about the convergence of objects, databases and XML. He mentioned a chapter of Tim Neward’s Effective Enterprise Java book where Ted recommends designing the data first. Ted started Effective Enterprise Java 2 days before Patrick was born so I haven’t been reading it. The issue with designing the data first is that typically, a developer is predisposed towards one of the three poles of data design (XML, objects or relational DB) that will color that design unintentionally.

I realized I still lean towards OO when Dare pointed out the fact that XML (actually, I should say XSD) doesn’t really need derivation. Because of my OO background, it took me a while to digest that concept. But since XML is just data without behavior, it doesn’t need polymorphism the way that objects do. Consider the following schema:

<xs:complexType name="ctAddress">
  <xs:sequence>
    <xs:element name="Street" type="xs:string" maxOccurs="1" minOccurs="1" />
    <xs:element name="City" type="xs:string" maxOccurs="1" minOccurs="1" />
    <xs:element name="State" type="xs:string" maxOccurs="1" minOccurs="1" />
    <xs:element name="ZipCode" type="xs:string" maxOccurs="1" minOccurs="1" />
    <xs:any namespace="##targetNamespace" />
  </xs:sequence>
</xs:complexType>

This is version one of the address complex type schema for some arbitrary web service. Over time, we realize that we want to be more global in our addressing schema, so we want to add a country element. Since not all of the other services we interact with will be updating to the new schema, we need to make country an option element (i.e. minOccurs=”0″). While we could use schema inheritance to do this, we could also just duplicate the existing elements and add a country element:

<xs:complexType name="ctAddress">
  <xs:sequence>
    <xs:element name="Street" type="xs:string" maxOccurs="1" minOccurs="1" />
    <xs:element name="City" type="xs:string" maxOccurs="1" minOccurs="1" />
    <xs:element name="State" type="xs:string" maxOccurs="1" minOccurs="1" />
    <xs:element name="ZipCode" type="xs:string" maxOccurs="1" minOccurs="1" />
    <xs:element name="Country" type="xs:string" maxOccurs="1" minOccurs="0" />
    <xs:any namespace="##targetNamespace" />
  </xs:sequence>
</xs:complexType>

What’s interesting is that even though these two schema types are not related by inheritance, I can still validate XML addresses against either schemas. In fact, XML addresses with and without addresses validate against both schemas! That’s difficult to model in a typing system where objects have a bound to a single specific type. Other types of changes can be introduced so that break validation to one of the schema but not the other. For example, if we changed the minOccurs of ZipCode to zero, all messages that validate to the first schema would also validate to the second, but the reverse would not always be true. This is like a IsA relationship in OO, but in the wrong direction (a base message “is a” derived message, but a derived message is not always a base message).

The upshot of all this is that I think my argument against XML Serialization as a general concept is strengthened. While it does work in many scenarios, I can easily build XML messages and XSD schemas that don’t cleanly conform to an OO typing system. Since the flexibility in XML is critical (that’s what I’m using for loosely coupled public interop interfaces), I know I don’t want my schema design to be constrained to the limited set of scenarios that are supported by XML Serialization.

Even though I said I’d try and lay off the Patterns and Practices blog entries for a while, I had a customer email me his thanks for posting these links. So I decided to update my site a little to add links to practices and application blocks that I think are a must read. They’re now listed down the left hand side along with .NET links, .NET tools and the DevHawk Navigation menu. As new practices and application blocks are published, I’ll be adding to this list as appropriate.

I didn’t notice right away, but there’s a new version of the Operating .NET Applications book. The original version was missing the chapter on Instrumenting .NET applications since the Enterprise Instrumentation Framework (EIF) hadn’t shipped yet. EIF is downloadable from MSDN Subscriber Downloads for all Enterprise and Universal subscribers. The new version of the book also contains information on Sizing and Capacity Planning.